Climate Finance∗

Stefano Giglio1, Bryan Kelly2, and Johannes Stroebel3

1Yale School of Management, NBER, and CEPR2Yale School of Management, AQR Capital Management, and NBER3New York University, Stern School of Business, NBER, and CEPR

October 26, 2020

Abstract

We review the literature studying interactions between climate change andfinancial markets. We first discuss various approaches to incorporating cli-mate risk in macro-finance models. We then review the empirical literaturethat explores the pricing of climate risks across a large number of asset classesincluding real estate, equities, and fixed income securities. In this context,we also discuss how investors can use these assets to construct portfolios thathedge against climate risk. We conclude by proposing several promising direc-tions for future research in climate finance.

Keywords: Climate Change, Climate Risk, Physical Risk, Transition Risk, ESG

1 INTRODUCTION

Climate change is one of the defining challenges of our time, with the potential toimpact the health and well-being of nearly every person on the planet. In addition,climate change poses a large aggregate risk to the economy and the financial system

∗Prepared for Annual Reviews (www.annualreviews.org). AQR Capital Management is a globalinvestment management firm, which may or may not apply similar investment techniques or meth-ods of analysis as described herein. The views expressed here are those of the authors and notnecessarily those of AQR.

1

(see, e.g., Litterman et al., 2020). The tools of financial economics, designed forvaluing and managing risky future outcomes, can therefore help society assess andrespond to climate change risk.

Starting with the seminal work of Nobel Laureate William Nordhaus in the 1970s,researchers have studied the interactions between climate change and the economy.Fossil fuels are a critical input to production, so economic growth increases green-house gas emissions. Those emissions induce climate change, and climate change hasa potentially large negative feedback effect on future economic activity. However,many important aspects of climate change economics that are financial in nature —such as the pricing and hedging of risks stemming from climate change, the aware-ness and attitudes of investors towards these risks, and the effects of climate riskson investment decisions — have received less attention in the literature. Indeed, itis only recently that researchers in financial economics have begun to explore thesequestions. This burst of research activity constitutes a new and quickly growing fieldthat we refer to as “Climate Finance.” In this review article, we summarize some ofits theoretical and empirical contributions. We are optimistic that this body of workwill grow in size and influence, and that climate finance will contribute to effortsacross the physical and social sciences to address the challenges of climate change.

Why study climate change through the lens of financial economics? First, riskand risk preferences play an important role in dictating the optimal policy responseto climate change. Second, financial markets are a primary vehicle for mitigatingand hedging climate risk. They mitigate climate risk by facilitating the flow ofinvestment capital toward “green” projects, and away from “brown” industries andfirms, as we transition to an environmentally sustainable economy. Examples of theclimate risk-mitigating role played by financial markets include financial innovationin “green bonds” and a ramp up in “climate-aware” mutual funds. Financial marketsalso provide a venue for hedging climate risk. Indeed, one of the most fundamentalfunctions of financial markets is the sharing and transferring of risks. While climaterisk is an aggregate risk, the heterogeneity in exposure to climate change acrossdifferent firms and regions provides valuable risk-sharing opportunities. Likewise,heterogeneity in adaptability and risk tolerance makes some investors better suitedto bear this risk than others.

Our review of the current literature is organized into two parts. In the first sec-tion, we discuss efforts to incorporate climate risk into macro-finance models. Thepioneering work of Nordhaus (1977) paved the way for thinking about the interac-tion of the physical process of climate change with the real economy. Early papers inthis literature — such as Nordhaus (1977, 1991, 1992) — focused on optimal climatechange mitigation, and worked in deterministic settings. As such, these papers did

2

not directly speak to the ways in which climate change affects asset prices and riskpremia. Subsequent work extends these models to incorporate different aspects ofrisk and uncertainty about climate change and its link to the economy. These at-tributes include the stochastic nature of physical and economic processes as well asuncertainty about models of these processes (see, for example, the work by Kolstad,1992, Manne et al., 1992, Nordhaus, 1994, Kelly & Kolstad, 1999, Nordhaus & Popp,1997, Weitzman, 2001, 2009, Lemoine & Traeger, 2012, Golosov et al., 2014). Muchof this literature has focused on the way risks and uncertainties affect optimal mitiga-tion policies and the “social cost of carbon.” More recently, the financial economicsliterature has explored the implications of these models for the prices and returns offinancial assets.

In the second part of this review article, we discuss the empirical literature thatexplores the pricing of climate risk across a large number of asset classes. Thisliterature considers the price effects of at least two broad categories of climate relatedrisk factors: physical climate risk and transition risk. Physical climate risk includesrisks of the direct impairment of productive assets resulting from climate change;transition risk includes risks to cash flows arising from a possible transition to a low-carbon economy. A central element of the research designs in these papers is thatassets are differentially exposed to these climate risk factors: for example, houseslocated near the sea are more exposed to physical climate risks, while coal companiesare more exposed to transition risks. Many papers then combine the differentialexposure of assets within an asset class with time-varying attention paid to climaterisk in order to understand how this type of risk is priced in asset markets. We reviewresearch that documents climate-related asset price effects in equity markets, bondmarkets, housing markets, and mortgage markets. We also discuss recent work thatshows how one can use financial assets to construct portfolios that hedge climatechange risks.

2 CLIMATE RISK AND ASSET PRICES: THE-

ORY

To introduce climate change risk into an economic model, the researcher must takea stand on the main sources of uncertainty associated with the processes of climatechange and the economy. Possible sources include: uncertainty about the futurepath of economic activity; uncertainty about the future evolution of the climate;and uncertainty about the various components of the model itself (e.g., the modelparameters that capture the interaction between the climate and the real economy).

3

Each source of uncertainty has a different effect on equilibrium prices and risk premia.To organize the discussion, we begin by providing a stylized framework that helps

highlight the distinction between the effects of uncertainty about the future path ofeconomic activity and uncertainty about the evolution of the climate process. Aswe show, this distinction has radically different implications for asset prices and riskpremia. The model we present here is a simplified version of the one introduced inGiglio et al. (2020). Like all integrated models of climate change and the economy,our model specifies both the dynamics of the economy and the physical climateprocesses. In the interest of parsimony, both processes are highly stylized versions ofthose considered in the physical sciences. Given our interest in asset prices, we alsospecify the preferences of investors. We assume a representative-agent Lucas-treeeconomy and directly specify the dynamics of equilibrium consumption growth aswell as the dynamics of climate change:

∆ct+1 = µ+ xt + Jt+1, (1)

xt+1 = µx + ρxt + φJt+1, (2)

λt+1 = µλ + αλt + νxt + χJt+1. (3)

Equations 1 and 2 describe the evolution of aggregate consumption growth, ∆ct. Jtis the only shock in this economy; we discuss a number of interpretations of it ingreater detail below. Jt directly affects consumption growth, but it also potentiallyaffects the other variables in the system. xt represents time-varying expected con-sumption growth, whose persistence is ρ and whose innovations are also driven byJt. The parameter φ captures the way the shock Jt affects the future path of con-sumption growth, and therefore plays an important role in capturing the dynamicsof consumption and, ultimately, the term structure of risk and risk premia. Thelast equation represents the dynamics in the conditional distribution of Jt. In anygiven period, Jt+1 could take one of two values: −ξ ∈ (0, 1) with probability λt, or0 with probability (1− λt). This probability itself is time-varying, according to thedynamics presented in equation 3: in addition to an autoregressive term, it dependson lagged measures of economic activity (xt), as well as on the current shock to Jt+1.Note that in this simplified framework, there is no uncertainty about the model orits parameters: so risk in this economy is entirely due to the shock Jt+1.

While this model is very stylized, it can help us compare and contrast two popularways in which the literature has modeled climate risks. The first approach empha-sizes uncertainty about the path of climate change as a direct source of risk for theeconomy. The second approach starts from the observation that the evolution ofthe climate and its damages is tightly linked to economic activity; in this approach,

4

uncertainty about climate damages primarily stems from uncertainty about the fu-ture path of economic activity. Of course, in practice both channels are likely tobe at play at the same time. We review each channel within the framework of ourone-shock model.

Uncertainty about the path of climate change. We begin by studying theclass of models in which the main source of uncertainty is about the future pathof climate change. The most representative set of papers in this area thinks aboutclimate risks in terms of a relatively low-probability catastrophic event that coulddramatically impact the economy: a “climate disaster” (Barro, 2013, Weitzman,2012, 2014, Wagner & Weitzman, 2015). Such a climate disaster is often motivatedby reference to a climate “tipping point.” In the model above, this view coincideswith interpreting Jt as the realization of the climate disaster. The parameter λtthen captures the conditional probability of the climate disaster, and ξ is the size ofthe disaster. The occurrence of the climate disaster also affects the future path ofthe economy, since it directly affects expected consumption growth xt. Specifically,when φ > 0, the occurrence of the climate shock reduces not only consumption im-mediately, but also future expected consumption growth (as in Bansal et al., 2016).When instead φ < 0, there is partial mean reversion after a climate shock. The lattercase has an especially interesting interpretation when modeling climate change: itcaptures the ability of the economy to adapt to climate change and rebuild someof the lost output at a relatively faster pace (see the emerging literature on adap-tation to climate change, e.g., Brohe & Greenstone, 2007, Deschenes & Greenstone,2011, Desmet & Rossi-Hansberg, 2015, Burke & Emerick, 2016, Barreca et al., 2016).Through the parameter χ, a climate disaster realization also affects future climaterisks λt. Finally, the model allows for feedback effects between climate change andthe economy. Climate risk affects the economy (when the climate disaster occurs),and economic activity affects climate risk through the effect of xt on λt (modulatedthrough the parameter α): when economic activity is high, climate risk increases.But when the climate shock materializes, consumption is low. In this model, thereis no additional source of uncertainty about the path of economic activity beyondwhat is due to the climate shock.

Uncertainty about the path of the economy. As the early literature in cli-mate economics points out, economic activity is itself a driver of climate change,so uncertainty about economic growth generates uncertainty about climate change.In this vein, the typical model embeds chains of events of the following form: theeconomy experiences a positive growth shock, pollution increases alongside output,

5

the increase in CO2 emissions accelerates climate change, which in turn acceleratesclimate-related damages on the economy (Nordhaus, 1977, 1991, 1992). In short,higher growth today associates with larger negative effects on future growth throughthe climate feedback channel. Conversely, negative shocks to economic activity (asrecently experienced during the COVID-19 crisis) instead results in less pollutionand less climate-related economic damage.

This model view can be represented in the one-shock model above by interpretingJt as a shock to economic activity. A standard specification for climate changedamages (e.g., Nordhaus & Boyer (2000) and Nordhaus (2008)) posits that they scaleup with consumption. Suppose first that climate damages are a constant fractionof consumption, that is, they are Qt = τCt. Then, equations 1 and 2 representthe dynamics of net consumption (that is, aggregate consumption net of the climatechange damages), Jt is a standard shock to economic activity, and equation 3 capturesthe time-varying distribution of economic activity. As discussed in Giglio et al.(2020), the case in which the climate-related output tax varies over time (Qt = τtCt,with τt time-varying) is also nested in these equations by appropriately changing thedynamics of xt.

Asset pricing implications. The discussion above highlights that equations 1–3 nest two common paradigms for embedding climate risks in economic models:uncertainty about climate disasters versus uncertainty about economic activity andits climate feedback effects. Once the physical processes of consumption, climatechange, and climate damages are determined, the model is completed by choosinga utility function for the representative agent. This implies a stochastic discountfactor for the economy, which in turn implies equilibrium asset prices. From here,this toy model can be enriched in many dimensions, for example by introducingheterogeneity, production, governmental climate policies, and realistic informationsets for agents inside the model.

We first review a number of asset pricing implications of the two models presentedabove. The conclusions in this section are derived under power utility, as in Giglioet al. (2020). Before focusing on the central differences between the two modelingapproaches, let us begin by pointing out a common implication. Whether uncertaintyis directly about the climate or indirectly about climate through economic activityand feedback effects, realized climate damages have negative effects on the sectorsof the economy exposed to that risk. Put differently, in both models, an asset withpositive exposure to climate risk will decline in value when an adverse climate shockoccurs, whatever the origin of that shock.

The two modeling paradigms, however, have starkly different implications for risk

6

premia associated with climate damages. We discuss implications for both the leveland the term structure of discount rates. A number of other forces that are noteasily captured by our stylized model – such as alternative preference specificationsand model uncertainty – can also affect risk premia, and we discuss these forces inmore detail below.

When the uncertainty emanates directly from the climate process itself (e.g., inclimate disaster models), climate damage tends to be unexpectedly high in timeswhen consumption is low because climate disaster realizations are a primary driverof reduced consumption. Assets that are positively exposed to climate risk — thatis, assets with low payoffs when climate damages are high — thus tend to requirepositive risk premia. On the other hand, assets that are negatively exposed to climaterisk — such as marginal mitigation and hedging investments that pay off primarilywhen climate damages are realized — will have negative risk premia since these assetsprovide an insurance against bad (high marginal utility) states of the world. This hasimportant implications for the appropriate discount rates used to value mitigatinginvestments. For example, consider an investment that reduces CO2 emissions byone ton today, the value of which is often referred to as the “social cost of carbon.” Inclimate disaster models, the social cost of carbon is relatively high, because payoffsto investments that mitigate climate damages are discounted at rates lower than therisk-free rate.

On the other hand, if the main source of uncertainty is instead about the pathof the economy, the climate risk premium has the opposite sign. In these models,climate damages are assumed to be larger when the economy is performing better.An investment that mitigates climate damages tends to pay off in times when con-sumption is high and marginal utility is low. Therefore, these investments carrypositive risk premia and should be discounted at rates above the risk-free rate. Inthese models, the associated social cost of carbon is then relatively low. The lowvalue of climate mitigation investments reflects the agent’s unwillingness to pay formitigation because it pays off in good times when marginal utility is low. Lemoine(2015) explains this channel as follows: “Under conventional damage specifications,the consumption losses due to climate change increase in the level of consumption.As a result, emission reductions increase future consumption by a larger amountwhen future consumption is otherwise high. This mechanical correlation between fu-ture consumption and the future consumption benefits of emission reductions makesemission reductions seem like an especially risky investment and therefore works toreduce the policymaker’s willingness to pay for emission reductions.”

In addition to the sign of the risk premium, the stylized model also also allows usto explore how the term structure of risk premia is affected by different assumptions

7

about the term structure of the riskiness of cash flows. Consider an investmentthat can mitigate the effects of climate change at some point in the future. As wediscussed above, this investment commands a negative risk premium if uncertaintystems directly from the climate path, and a positive risk premium if uncertaintystems from the path of economic activity. But how do these premia vary alongthe term structure (i.e., along the horizon at which the benefits of the investmentmaterialize)? The answer depends on the dynamics shocks in the model.

First, we assume that shocks to expected growth are persistent, ρ > 0, as inmost dynamic macro-finance models. Next, if φ > 0, there are long-run risks inconsumption, since today’s shock J also affects consumption growth in the futurein the same direction. In that case, long-term assets are more exposed to climateshocks than short term assets. Long-term discount rates are then larger (in absolutevalue) than short-term discount rates. If the primary source of uncertainty is aboutthe climate path, risk premia for climate mitigation investments are negative andbecome even more negative at longer horizons. If the primary uncertainty is aboutthe path of economic activity, discount rates are positive in the short term andbecome increasingly positive with the horizon.

Alternatively, if φ < 0, then the economy recovers quickly from a negative shockJ , growing faster than average during the recovery. In that case, long-term assetsbenefit from a recovery in a way that short-maturity assets do not. Because ofthis, the future is less risky, and risk premia revert toward zero as horizon increases.If uncertainty is about the climate path, risk premia for climate mitigation startsnegative but increase with the horizon. If uncertainty is about the path of economicactivity, risk premia are positive at short horizons but decrease with the horizon.

In our stylized model with power utility, the level and term structure of discountrates are only driven by different assumptions about the term structure of the cashflow risk. As a result, empirical evidence on the term structure of discount rates ofassets exposed to climate risk allows researchers to discipline the cash flow processesin equations 1–3. In particular, recent advances have been made in estimating theterm structure of discount rates for equities (Binsbergen et al., 2012, Van Binsbergen& Koijen, 2017, van Binsbergen et al., 2013) and real estate (Giglio et al., 2015, 2020),both of which are asset classes that are exposed to climate risk (see the next section).For both asset classes, researchers have found positive risk premia that decline withthe horizon. A central insight from Giglio et al. (2020) is that these patterns speak infavor of a version of our model along the “climate disaster” lines, where the primarysource of uncertainty is the path of climate change, combined with partial meanreversion of the economy following a climate shock (φ < 0), perhaps because of anability of the economy to adapt to the new climate.

8

To sum up, the debate around the term structure of discount rates for valuinginvestments to mitigate climate change (and its effects on the social cost of carbon)can in large part be traced to different assumptions about the nature of the shocksthat mitigation investments are hedging, and about the dynamics of the economyand the climate in response to those shocks. While this two-dimensional distinctiondoes not fully span the variety of models that have been written in the literature,it helps to understand what has lead the literature to reach different (sometimesopposite) conclusions.

So far we have focused on discussing the asset pricing implications of climatechange that result from different specifications of the shocks and of the dynamicsof the processes. However, researchers have explored additional forces that alsoaffect risk premia and their term structures, such as preferences and uncertainty. Ashighlighted by a recent literature, drawing meaningful conclusions on the level andterm structure of risk premia for climate investments requires taking a stand on themodel in its entirety (see Gollier & Weitzman, 2010, Traeger, 2013, Arrow et al.,2013). We next discuss in turn how preferences and uncertainty affect climate riskpremia.

Preferences. Preferences for time and risk naturally play an important role indetermining discount rates and their term structure.

First, it is important to establish the rate of pure time preference. Stern (2007)took the view that ethical considerations should dictate this rate, and suggestedusing a pure time preference coefficient of effectively zero, giving the same weightto all generations. Most of the literature, such as Nordhaus (2007) and Weitzman(2007), has instead argued for using rates of pure time preference that are greaterthan zero, consistent with observed savings and investment behavior (see also thediscussion in Arrow, 1995).

In addition, the climate finance literature has explored a number of alternativerisk preferences. The most prominent alternative to power utility is Epstein–Zinutility (see, for example, the work of Gollier, 2002, Crost & Traeger, 2014). AnEpstein–Zin investor’s marginal utility depends not only on the one-period innovationin consumption growth (as in the power utility case that we described above) butalso on news about consumption growth at future horizons. This specification ofpreferences has two main consequences for thinking about climate risks (for suitableparameterizations that induce a preference for early resolution of uncertainty): first,it affects the level of climate risk premia, because it amplifies the utility consequencesof climate shocks if they have long-term implications for the agent’s consumptiongrowth (as they do, for example, in Bansal et al., 2019). Second, they affect the term

9

structure of discount rates, as well as the optimal timing of mitigation investments.For example, Daniel et al. (2019) show that Epstein–Zin preferences increase thevalue of mitigating climate change early, and result in declining paths of CO2 pricesover time, as uncertainty gets resolved.

Model Uncertainty. The vast majority of asset pricing theory is formulated underthe assumption of rational expectations, so that investors inside the model knowthe exact probability laws that govern their decision making environment. Theillustrative model introduced above is an example of this modeling approach. Butin the context of climate change (and likely more generally), rational expectationsendow investors with an unrealistic understanding of their environment. As Hansen(2014) emphasizes, “it is often not clear what information should be presumed onthe part of economic agents, how they should use it, and how much confidence theyhave in that use.” Put differently, in rational expectations models, investors face theuncertainty about stochastic realizations from a known probability law. In reality,investors also wrestle with ambiguity, or uncertainty about the true probability law.Nowhere could this be truer than in economic models of climate risk. Is it implausiblethat economic agents know with any degree of certainty the precise nature or severityof climate risks that are facing them, a topic of substantial disagreement even withinthe scientific community. As Lemoine (2020) aptly notes,

“Uncertainty is fundamental to climate change. Today’s greenhouse gasemissions will affect the climate for centuries. The emission price thatinternalizes the resulting damages depends on the uncertain degree towhich emissions generate warming, on the uncertain channels throughwhich warming will impact consumption and the environment, on theuncertain future evolution of greenhouse gas stocks, and on uncertainfuture growth in productivity and consumption.”

Recent climate-economic theory begins to confront ambiguity in climate economicmodels using tools from the theory of decision under uncertainty. Brock & Hansen(2018) and Heal & Millner (2013) lay out the general argument for, and technicalapproaches to, featuring model uncertainty in climate economic models. Heal &Millner (2013) catalogs the sources of ambiguity as “scientific uncertainty” (e.g., theeffect of atmospheric CO2 concentrations on surface temperatures) versus “socio-economic uncertainty” (e.g., even if we knew the model for the climate effects ofCO2, we remain uncertain about how societies react to these effects). Surveyingpredictions from a wide range of climate research and meta-analyses, Brock & Hansen(2018) illustrate wide heterogeneity in forecast distributions among leading climate

10

Figure 1: Temperature Forecasts From Various Climate Models

Source: Climate Lab Book (http://www.climate-lab-book.ac.uk/).

models. As an example, Figure 1 shows the divergence in forecasted temperatureanomalies across several models and across parameter values within those models.

In models of ambiguity, investors typically assign probability weights to models ina Bayesian fashion based on a combination of their prior model weights and the datathey have observed. Investors make savings and consumption decisions based on thecompound uncertainty they face based on their posterior weights over models andthe stochasticity within each model. Preferences interact with ambiguity throughrecursive preferences or, more directly, through robust or ambiguity averse prefer-ences. Brock & Hansen (2018) outline a range of potential modeling choices for thespecific forms that model uncertainty might take and for the preference or decisiontheoretic constructs that agents in the model use to make savings and consumptionchoices in the face of uncertainty.

Lemoine (2020) argues that accounting for model uncertainty leads to higherestimates of the social cost of carbon than would otherwise prevail. In modelswith recursive preferences, uncertainty (especially with regards to economic dam-ages arising from temperature increase) raises the social cost of carbon by an orderof magnitude relative to a model with the same preference structure but no ambigu-ity. Parameters that govern climate-related economic damages are random variableswhen approached from a model uncertainty viewpoint. Uncertainty thus introducesa new channel that impacts asset prices in the form of covariance between modelparameters and agents’ consumption. This induces precautionary savings and risk

11

premia effects in addition to those resulting from stochastic shocks in standard un-ambiguous models. Viewing damage uncertainty as a compound lottery, when theagent “draws” an especially adverse damage parameter, carbon mitigation becomesespecially valuable and raises the social cost of carbon (as long as relative risk aver-sion is greater than one, as commonly assumed in calibrations of macro and financemodels).

In the face of model uncertainty, Bayesian economic agents gradually update theirbeliefs about their environment based on the arrival of new data. Since Bayesianposteriors follow a martingale, these updates constitute permanent shocks from theperspective of the learner. As Johannes et al. (2016) and others point out, thesebelief update shocks are especially risky for agents with recursive preferences and, asa result, risk premia are magnified relative to baseline models with no ambiguity orwith non-recursive utility. This “learning as a long-run risk” perspective motivatesthe measurement of climate risk through news arrival in the empirical analysis ofEngle et al. (2020) discussed in Section 3. The persistent effects of belief updatingleads model uncertainty to accumulate over time, partially offsetting the effects ofdiscounting on damages in the distant future. This mechanism is at the core of thelarge increase in the social cost of carbon in Lemoine (2020).1

Once we acknowledge that agents in a climate model face ambiguity, it is natu-ral to consider decision theoretic frameworks beyond subjective expected utility inorder to accommodate the well documented human tendency of ambiguity aversion(Ellsberg, 1961). As a leading example, Barnett et al. (2020) analyze the social costof carbon under model uncertainty while accounting for ambiguity aversion. In theirsetup, ambiguity takes the form of multiple potential models of the impact of car-bon emissions on temperature and economic damages caused by high temperatures.Agents make decisions amid this ambiguity based on the recursive smooth ambigu-ity aversion preferences of Hansen & Miao (2018). While the comparative statics inLemoine (2020) analyze incremental carbon costs associated with model uncertaintyin an expected utility framework, Barnett et al. (2020) analyze the additional incre-mental effects of ambiguity aversion on the social cost of carbon. Holding fixed theextent of model uncertainty, they compare model calibrations with ambiguity averseinvestors versus a model with ambiguity neutrality. Ambiguity aversion magnifies

1A similar model analyzed by Dietz et al. (2018) calibrates a comparatively small cost of car-bon, arguing that “the positive effect on the climate beta [covariance of consumption and themarginal social cost of emissions] of uncertainty about exogenous, emissions-neutral technologicalprogress overwhelms the negative effect on the climate beta of uncertainty about the carbon-climate-response.” Lemoine (2020) chalks this difference up to the larger model uncertainty used in hiscalibration compared to that of Dietz et al. (2018).

12

the cost of carbon by roughly 60% to 70% in current value terms relative to thebaseline scenario with model uncertainty but ambiguity neutrality.2

3 CLIMATE RISK ANDASSETMARKETS: EM-

PIRICAL EVIDENCE

In recent years, the finance literature has made substantial progress in understandingthe effects of climate change on asset prices across a number of asset classes. Thisresearch effort has the immediate benefit of helping us quantify the forces that linkthe climate and the economy, and therefore allows us obtain a better understanding ofthe underlying structural relationships discussed in the previous section. In addition,we will show that understanding the empirical relationship between climate changeand asset prices has the benefit of giving us implementable indications on how to usefinancial markets to hedge climate risks.

For researchers interested in understanding the effects of climate change on assetprices, it is important to note that there are actually several different categoriesof climate risks potentially priced in asset markets, and that these different risksoften do not materialize at the same time. Broadly speaking, climate risks can bedivided into physical risks and transition risks. The physical risks of climate changeare those that result directly from the effects of changes in the climate on economicactivity. For example, the threat of damage from rising sea levels to firms’ productionfacilities close to the sea, and the associated destruction of real estate values, wouldbe considered a physical climate risk. Transition risks cover a wide range of effectson firms’ operations and business models that come from a possible transition to alow-carbon economy. One example of a transition risk is the possible introduction ofa carbon tax that might leave fossil fuel companies with stranded assets that are nolonger profitable to operate. In addition to such regulatory risks, transition risks alsoinclude technological advances and changing consumer preferences away from high-carbon activities. While realizations of physical and transition risks need not occur atthe same time, they are often correlated, and might even move in opposite directions.For example, the introduction of a carbon tax — a realization of negative transition

2Barnett (2017) also pursues this line of research to understand asset prices and production whenagents have both recursive preferences and aversion to uncertainty about the climate model. Heanalyzes a dynamic general equilibrium model where production involves a mix of cheap, pollutingfossil fuels and expensive, clean green energy. Accounting for aversion to climate model misspecifi-cation increases the market price of climate risk in the model by as much as an order of magnituderelative to a model with no ambiguity aversion.

13

risk — might reduce the likelihood of future negative realizations of physical climaterisks.

Different assets may be positively or negatively exposed to these types of climaterisks; in other words, realizations of both physical and transition risks will have win-ners and losers in asset markets. For example, while coal companies would likelysuffer from realizations of transition risks, renewable energy companies might bene-fit. And while climate change will negatively affect the value of coastal real estate, itmight also increase the value of farmland in colder regions of the world. Given thesedifferent risk categories, and the different exposures of various assets to these riskcategories, one of the important common challenges for all approaches to exploringhow climate risks affect asset markets is to obtain measures of different assets’ ex-posures to both physical and climate risks. Our discussion below will highlight howdifferent researchers have approached this challenge.

A second important challenge to documenting how climate change is priced inasset markets is that the rise in investor attention to climate risk is a fairly recentphenomenon. As a result, while climate risk may be priced in asset markets today,it might not have been 10 or 15 years ago. This reduces the ability of researchers toexploit time series variation in documenting how climate change affects asset prices.In addition, the availability of only a short time series makes it hard to estimate theclimate risk premium, which, as we saw in the previous section, plays a fundamentalrole in distinguishing between theories. As a result, our understanding of the pricingof climate risks in financial and non-financial assets is likely to evolve substantiallyover the coming years, as more time series data become available.

3.1 CLIMATE RISK AND FINANCIAL ASSETS

A number of papers have explored the extent to which climate risks are priced infinancial assets. This literature often starts from the observation that a growingnumber of large institutional investors have declared environmental, social, and cor-porate governance (“ESG”) sustainability an important objective in their portfolioallocation process. Naturally, climate change is a focal issue in ESG investing. Forexample, Larry Fink of BlackRock wrote in his 2020 letter to CEOs that “our invest-ment conviction is that sustainability- and climate-integrated portfolios can providebetter risk-adjusted returns to investors.” He concludes that, in response to climatechange, “in the near future – and sooner than most anticipate – there will be asignificant reallocation of capital.” This statement is consistent with the systematicanalysis of presented in Krueger et al. (2020). These authors conduct a survey ofactive investment managers to explore their approaches to managing climate risk.

14

They find that investors believe that climate change has significant financial implica-tions for portfolio firms, and that considerations of climate risk are important in theinvestment process. For example, 39% of investors in the survey reported to be work-ing to reduce the carbon footprints in their portfolios. These survey responses arealso consistent with findings from Alok et al. (2020), who show that fund managersadjust their portfolios in response to climatic disasters. Pedersen et al. (forthcoming)provide an ESG CAPM framework and outline how investor beliefs and preferencesregarding climate change risks (and ESG considerations more broadly) fit in withthe factor model paradigm that dominates empirical asset pricing research.

Equity Markets. Given the attention that investors dedicate to climate change,a growing literature explores the pricing of various dimensions of climate risk inequity markets (e.g., Hong et al., 2019). Much of this literature has focused on theeffects of regulatory climate risk, where different measures of carbon intensity orenvironmental friendliness are often used as proxies for regulatory climate risk.3 Forexample, Bolton & Kacperczyk (2020) analyze U.S. equity markets, and demonstratethat firms with higher carbon emissions are valued at a discount. Quantitatively, theauthors estimate that a one standard deviation increase in emissions across firms isassociated with a rise in expected returns of roughly 2% per annum. The authorstrace this effect at least in part to exclusionary screening performed by institutionalinvestors to limit the carbon risk in their portfolios. In related work, Hsu et al. (2020)show a similar spread in average returns between high- and low-pollution firms, andlink it to uncertainty about environmental policy. Engle et al. (2020) document thatstocks of firms with high E-Scores — which the authors argue capture lower exposureto regulatory climate risk — have higher returns during periods with negative newsabout the future path of climate change. Similarly, Choi et al. (2020) explore globalstock market data and find that stocks of carbon-intensive firms underperform duringtimes with abnormally warm weather, a period when investors’ attention to climaterisks are likely to be particularly high. Barnett (2020) uses an event study analysisto explore financial market impacts of regulatory risk. He finds that increases inthe likelihood of future climate policy action lead to decreased equity prices forfirms with high exposure to climate policy risk. Similar evidence of the pricing ofclimate risk can be found in equity options markets. Ilhan et al. (2019) show thatthe cost of option protection against extreme downside risks is larger for firms withmore carbon-intense business models, and particularly so at times when there is anincreased public attention to climate risk.

3A number of recent papers have attempted to provide new measures of firm-level exposures toclimate risk that are based, for example, on the language around climate change used in earningscalls (Sautner et al., 2020, Li et al., 2020).

15

In addition to climate change affecting firm valuations through the exposure oftheir business model to various forms of climate risk, investors also appear to rewardfirm efforts to mitigate these risks. For example, Perez-Gonzalez & Yun (2013) studythe effects of managerial efforts to mitigate firms’ climate risk exposures on thosefirms’ valuations. Based on firm-level weather hedging data (using CBOE weathercontracts), they show that climate hedgers have higher valuations, and that thiseffect is more pronounced for more climate sensitive firms. This is consistent withfindings from the survey in Krueger et al. (2020), which highlighted that investorswere actively engaging with management — both informally and formally throughshareholder proposals — to work with them to reduce their climate risk exposures.

Fixed Income Markets. Climate risks may also affect financial assets beyondequities. Municipal bond markets are a particularly interesting setting for analyzingthe financial market implications of climate risk. In particular, when considering thephysical risks of climate change, firms may be at risk depending on the location oftheir production facilities. However, even the most exposed firms usually have theoption of relocating their modes of production to other geographies. Municipalitieshave no such luxury. As a result, one would expect that municipal debt backed bytax revenues from localities more exposed to physical climate risks such as rising sealevels or wildfires would trade at a substantial discount. In evidence along these lines,Painter (2020) shows that at-issuance municipal bond yields are higher for countieswith large expected losses due to sea level rise (SLR). Consistent with the hypothesisthat such price differences reflect the pricing of climate risk, he finds that this effectis concentrated in long-dated bonds and essentially absent at short maturities overwhich the likelihood of SLR remains low. In related work, Goldsmith-Pinkham et al.(2019) show via a structural model that this effect of SLR on municipal bond yieldsis tantamount to a 3–8% reduction in the present value of local government long-runcash flows.

Climate risk may also be priced in other fixed income markets, such as the cor-porate bond markets studied by Huynh & Xia (2020). These authors first calculatethe covariance of each bond’s returns with the climate news index constructed byEngle et al. (2020). Bonds with a more positive covariance are those that performrelatively well when bad news about climate change emerges, suggesting they areweakly or even negatively exposed to realizations of climate risk. Consistent withclimate risk being priced in corporate bond markets, Huynh & Xia (2020) find thatthose positive-covariance bonds have lower returns.

Another class of fixed income assets with valuations that might be affected byclimate risk are “green bonds” whose proceeds are expressly linked to environmen-tally friendly projects (e.g., renewable energies, clean transportation, etc.). Baker

16

et al. (2018) study a sample of more than 2,000 municipal and corporate green bondsand find that green bonds trade at lower yields than bonds with similar attributesthat lack a green designation. Quantitatively, after-tax yields at issue for greenbonds are roughly 6 basis points below yields paid by otherwise non-green equivalentbonds. The authors attribute these pricing differences to a subset of investors thathave a nonpecuniary component of utility, such as a sense of social responsibilityfrom holding green bonds. Realizations of climate risk may move the magnitude ofthis nonpecuniary valuation component, and will thus affect the valuations of thesebonds.

Hedging Climate Risks Using Financial Assets. The research describedabove shows that climate change is a significant risk factor determining asset prices.This observation then invites the question of how investors can mitigate the risks thatclimate change poses to their portfolios. This is particularly important since manyof the effects of climate change are sufficiently far in the future that neither financialderivatives nor specialized insurance markets are available to directly hedge thoselong-horizon risks. Instead, investors are largely forced to insure against realizationsof climate risk by building hedging portfolios on their own.

Engle et al. (2020) propose an approach to hedging climate risk that combinestraditional dynamic trading arguments from financial theory with novel statisticalmeasurements using textual analysis. The paper’s first insight follows the Black &Scholes (1973) and Merton (1973) logic that a dynamic strategy to hedge graduallyarriving news about future climate change can approximately replicate an infeasiblecontract that directly pays off in the event of a future climate disaster. The keyquestion, then, is: how to measure such news? To do this, Engle et al. (2020) extracta “climate news” index based on coverage of climate change by The Wall StreetJournal (WSJ). The index — which is available to other researchers, and which hasbeen used by Huynh & Xia (2020) and others — measures the extent to which WSJarticle text overlaps with climate change discourse in authoritative texts publishedby various governmental and research organizations. The index, shown in Figure 2,displays intuitive variation over time. The level of climate news coverage graduallyrises over time and spikes around topical global climate events. In their baselineapproach, Engle et al. (2020) interpret rising coverage of climate related topics asthe arrival of bad news about future climate change. They validate this approach bycomplementing their WSJ-based analysis with additional sentiment-based studies ofclimate coverage in newspapers.

With this measure of the arrival of climate news, Engle et al. (2020) propose anapproach to dynamically hedging climate news based on factor mimicking portfolios.Specifically, the authors use relatively easy-to-trade basis assets (U.S. equities) to

17

Figure 2: Climate Change News Index

1985 1989 1993 1997 2001 2005 2009 2013 2017DATE

25

50

75

100

125

150

175

200

225

WSJ

NOAA produceseducational film aboutglobal warming"The Climate Factor"

IPCC Formed

Adoption ofUNFCCC

UNFCCC meetsTreatyRequirement

Kyoto ProtocolAdopted

Bush withdraws fromKyoto Protocol

Kyoto ProtocolEffective

4th IPCC Report

G8 Climate Agreement

2009 CopenhagenUN Climate ChangeConference

2012 DohaUN Climate ChangeConference

3rd National ClimateAssessment &EPA Climage ChangeInitiative

ParisAgreement

TrumpWithdraws fromParis Agreement

Source: Engle et al. (2020).

mimic the WSJ climate news index. Intuitively, the approach is to systematicallyown or overweight stocks that rise in value when (negative) news about climatechange materializes and likewise short or underweight stocks that fall in value on thearriveal of this news. In doing so, the hedge portfolio profits when adverse climatenews hits. The authors show how to continually update the hedge portfolio usingevolving information about which stocks are most susceptible and which are mostresilient to climate risk.

To implement this dynamic hedging strategy, it is necessary to determine whichfirms increase or decrease in value when there is news around climate change. Engleet al. (2020) solve this problem by proxying for firms’ climate risk exposures using“E-Scores” that capture various aspects of how environmentally friendly a firm is.The hedge portfolio would then overweight high-E-Score firms, and underweight low-E-Score firms, with the relative weights updated dynamically as more data on therelationship between E-Scores, climate news, and asset prices is obtained. While it isstraightforward to construct such a hedge with the benefit of hindsight, the true testof a hedge portfolio is its ability to profit in adverse conditions on an out-of-samplebasis. Indeed, Engle et al. (2020) find an out-of-sample correlation of 20% to 30%between the return of the hedge portfolio and innovations in the WSJ climate change

18

news index. In summary, the paper provides a rigorous methodology for constructingportfolios to hedge against climate risks that are otherwise difficult to insure.

It is worth pointing out an additional connection between this empirical approachand the theories discussed in the previous section. If uncertainty about future climatedamages (for example, about the model parameters) is resolved slowly over time, thenthe news about the future path of the economy will itself represent a risk factor forthe investor. A portfolio that hedge the news about future climate, like the onebuilt by Engle et al. (2020), would then be useful not only to dynamically hedgethe long-term realization of the damages from climate change, but also to hedgethe risks represented by the arrival of information over time (resolution of modeluncertainty). This provides an additional advantage of this hedging portfolio (onethat would be particularly valuable for investors, like Epstein–Zin investors, that areespecially averse to such news shocks).

Finally, mimicking portfolios are closely related to risk premia. As is well knownfrom the asset pricing literature, the risk premium of any nontradable risk (likeclimate change risk) is the expected excess return of the corresponding hedging port-folio.4 Once the hedging portfolio for climate risks is built, therefore, one couldestimate the climate risk premium by looking at the time-series average return ofthis portfolio. As highlighted above, however, this exercise is difficult with a shorttime series. For example, suppose that the true climate risk premium is positive,as many of the models reviewed in the previous section imply. The climate-hedgingportfolio should then have a negative risk premium: its average return should be, inthe long run, below the risk-free rate. However, if one computes the average returnof this hedging portfolio over the last 10-15 years, one might as well find a positiveaverage return, because this short time period is dominated by the realization of aseries of bad news about climate change (see Figure 2). This would then lead toestimating the wrong sign for the climate risk premium. Given the short time se-ries available, it is important to keep these caveats in mind when interpreting thehistorical average returns of climate-exposed portfolios.

3.2 CLIMATE RISK AND HOUSING AND MORTGAGEMARKETS

A growing literature has started to explore the effects of climate change on real estatevaluations, and through this channel, on the mortgage market. This focus on realestate is hardly surprising. Indeed, since the value of real estate is tightly linked to

4The hedging portfolio can be constructed either directly, via mimicking-portfolio regressions,or indirectly, for example using two step cross-sectional regressions.

19

the value of the land it is built on, it is natural to suspect that physical climate riskfactors, such as rising sea levels and wildfires, might directly affect real estate prices.Take rising sea levels as an example: Hauer et al. (2016) estimates that a 1.8 metersea level rise by the year 2100 would affect about 13.1 million Americans. Similarly,Zillow economist Krishna Rao (2017) calculates that a six feet sea level rise wouldput 1.9 million homes worth about $882 billion at risk of flooding, with about halfthe losses coming from Florida alone. While many of these damages might only arisesome decades in the future, the low long-run discount rates for real estate discussedin the previous section mean that present-day real estate prices might already besignificantly affected by climate risk.

To explore whether or not climate change risk is priced in real estate today, onewould ideally want to compare the valuations of two otherwise identical propertiesthat are differentially exposed to physical climate risk factors such as rising sealevels, flooding, hurricanes, or wildfires. The challenge with this empirical exerciseis that all houses are, by nature, a unique combination of aspects of location andstructure (Kurlat & Stroebel, 2015, Piazzesi et al., 2020), which complicates thesearch for comparable units that might be differentially exposed to climate risk.More problematic still, some of the locational or structural aspects that contributeto the value of a particular property might correlate with the property’s exposure toclimate risk. To give a concrete example: properties with beach access are likely tobe more exposed to rising sea levels than properties that are located further inland.And, since beach access is a valuable amenity, beach front properties will usuallytrade at a premium relative to other properties. This does not mean, however, thatclimate risk is not priced. Instead, it means that the particular location of theproperty has both a positive amenity value, which increases the flow utility from theproperty, and is associated with higher climate risk, which affects the realization andvaluation of future cash flows. While hedonic regressions allow researchers to controlfor observable differences in quality features of properties, such as the property’ssize, many important location features are hard to measure and therefore difficult tocompletely control for in a regression.

Many of the researchers studying whether climate risk is priced in real estatemarkets therefore exploit additional sources of variation, often time-series variationin climate risk, beyond the across-property variation in climate risk exposures. Theidentifying assumption in these analyses is the following: as long as the amenityvalue of beach access does not change when climate risk changes, differential pricechanges of more-exposed and less-exposed properties can be informative about thepricing of climate risk in housing markets. However, such an analysis presents asecond challenge: true climate risk is a relatively slow-moving object that does not

20

provide much of the high-frequency time-series variation required to identify howit is priced using the approach just described. Researchers have therefore tried toexploit time-series variation in the attention to climate risk in the housing market.Indeed, even though true climate risk might not change much from year to year, theextent to which housing market participants focus on these risks changes much morefrequently, and one would thus expect the pricing implications of climate risk to beparticularly strong when households pay more attention to these risks

A recent paper using a research design along these lines is Giglio et al. (2020).They explore the pricing of the risk of rising sea levels in four coastal U.S. states— Florida, New Jersey, North Carolina and South Carolina — that are particularlyexposed to this physical climate risk factor. To measure different properties’ exposureto climate risk, the authors geo-code the addresses of all properties bought andsold in these states between 2008 and 2017. They then map the property locationsto information provided by the National Oceanic and Atmospheric Administration(NOAA) that indicates which regions will be flooded should sea levels rise by sixfeet or more. The authors document substantial variation in exposure to climaterisk across properties in the same narrow geography. This variation is driven, forexample, by differences in elevation, and allows the authors to compare the pricingof climate risk while holding local housing market conditions fixed.

To identify the pricing of climate risk, Giglio et al. (2020) also construct a mea-sure of attention paid to this risk in housing markets. To construct this measure,which varies across both time and space, the authors analyze the textual descriptionsof properties in the universe of for-sale and for-rent property listings from Zillow, amajor online real estate listings service. The resulting “Climate Attention Index”is constructed by calculating the proportion of for-sale listings with property de-scriptions that contain climate risk-related words and phrases such as “hurricanes”,“FEMA”, “floodplain”, and “flood risk.”Most of the flagged listings include descrip-tions that highlight that a specific property is less exposed to climate risk (e.g., “Notin a flood zone, it’s high and dry!”). This is unsurprising: if you are selling a housethat is not exposed to climate risk, this is something worth highlighting in a propertylisting, in particular in areas and at times when potential buyers pay more attentionto these risks. In terms of spatial variation, more attention to climate risk is paidin coastal regions, which have a higher average exposure to these risks. In the timeseries, attention to climate risk increases after salient natural disasters. For example,the Climate Attention Index in New Jersey increases substantially between 2011 and2013, around the time of Hurricane Sandy, which rendered 20,000 homes in the stateuninhabitable.

Using these data, Giglio et al. (2020) show that while properties in a flood zone

21

generally trade at a premium compared to otherwise similar properties (likely becauseof positive amenities such as beach access), this premium compresses in periods withelevated attention paid to climate risk. Quantitatively, a doubling in the ClimateAttention Index (i.e., a doubling in the share of listings that mention climate risk-related words) is associated with a relative 2.4% decline in the transaction prices ofproperties in the flood zone.

One possible concern with these results is that they might not just capture thepricing of future climate change risk. Instead, these estimates might also pick upchanges in the flow-utility of climate risk-exposed properties that could be correlatedwith climate risk attention. For example, as discussed above, it appears to be thecase that climate risk attention rises after hurricanes; if those hurricanes also havea particularly strong direct effect on the utility of living in properties located inflood zones, this might explain the relative price decline of these properties. Giglioet al. (2020) rule out such a story, by showing that the relative rents of propertiesmore exposed to climate risk do not decline when attention to climate risk increases.Since any decline in the flow-utility from exposed properties that is correlated withincreases in climate risk attention would also affect these rental units, this findingemphasizes that the relationship between climate risk attention, climate risk expo-sure, and price is driven by changes in the expected realization and valuation offuture cash flows (rents).

A number of other papers exploit related research designs to explore the pricing ofclimate risk in real estate markets. Bernstein et al. (2019) also explore the relation-ship between house prices and sea level rise (SLR). They find that houses that areexposed to sea level rise sell for a discount compared with observably equivalent un-exposed properties. The authors are able to control for the distance from the beach,which allows them to alleviate some concerns around differential amenity values ofthese properties. Quantitatively, properties that will be inundated after one foot ofglobal average SLR sell at a 14.7% discount, properties inundated with two to threefeet of SLR sell at a 13.8% discount, and properties inundated with six feet of SLRsell at a discount of 4.4%. Baldauf et al. (2020) present related evidence suggestingthat the extent to which physical climate risk is priced in housing markets dependson whether the local population believes in climate change. Bakkensen & Barrage(2017) explore a similar point, highlighting that when individuals who do not believein climate change disproportionately sort to purchase more exposed properties, thiswill reduce the extent to which climate change risk is priced in housing markets.5

5Using similar data but a different identification method, based on cross-sectional differencesin relative sea level rise due to vertical land motion, Murfin & Spiegel (2018) find instead a muchsmaller effect of climate change exposure on house prices, highlighting the importance of the iden-

22

A related set of papers has explored the effect of hurricanes on house prices,arguing that a recent hurricane makes future costs of climate change more salientto individuals, even in cases where the Hurricane did not affect a particular prop-erty. Ortega & Taspinar (2018) show that following Hurricane Sandy there was asignificant and permanent relative price decline of New York City properties in floodzones, even if they were not damaged by Sandy. Gibson et al. (2017) also find thatincreasing salience of climate risk following hurricanes reduces the relative valuationof more exposed properties in the New York housing market. Eichholtz et al. (2019)find similar results in the commercial real estate market. The authors study trans-actions in three cities, New York, Boston, and Chicago, before and after the shiftin the salience of flood risk caused by Hurricane Sandy. They find that propertiesexposed to flood risk experience slower price appreciation after the storm than equiv-alent unexposed properties. As the previous papers, Eichholtz et al. (2019) concludethat the price effect is persistent and not driven by physical damage incurred fromHurricane Sandy.

While rising sea levels are a first-order concern for coastal home owners, theyare not the only channel through which climate change poses a physical risk forreal estate values. Another salient risk of climate change is the increasing danger ofwildfires in many states. For example, Corelogic (2019) found that nearly 776,000homes with an associated reconstruction cost value of more than $221 billion were atextreme risk of wildfire damage. Many of the most at-risk properties are in CaliforniaMSAs, but properties in Texas and Colorado are also potentially at risk. Garnache& Guilfoos (2019) explore whether such wildfire risk is priced, and find that the priceof homes drops when these home are designated to be in a wildfire risk zone (see alsoMcCoy & Walsh, 2018).

Since most residential real estate is purchased with a mortgage, climate risk willalso affect the valuations of these mortgages. Consistent with this, recent researchhas shown that realizations of wildfire risk and flooding lead to increased mortgagedefault (Issler et al., 2019). Similarly, recent evidence provided by Ouazad & Kahn(2019) suggests that perceived climate risk affects banks’ decisions to securitize orig-inated mortgages (see also Keenan & Bradt, 2020). This suggests that the federalgovernment, through the default-risk guarantees that Fannie Mae and Freddie Macprovide to agency-insured MBS, may be directly exposed to the effects of climaterisks in real estate markets.

tification assumptions for the conclusions in this branch of research.

23

4 CLIMATE FINANCE: A RESEARCHAGENDA

Climate change will be a first-order issue facing our society for many years to come.Researchers in financial economics have only recently turned their attention to ex-ploring the many ways in which climate change will affect financial markets. Muchprogress has been made on both modeling the relationship between climate, the econ-omy, and asset prices as well as on documenting the many ways in which climaterisk is already priced in financial markets. But much work remains to be done.

On the modeling side, advances in computational power will allow researchers tomodel the various feedback loops between climate change and the real economy withincreasing sophistication. While the basic economic mechanisms of these models willbe similar to those discussed in this review, such modeling advances will provide newand improved quantifications of important objects such as the social cost of carbon.

On the empirical side, there is substantial scope for improvements of the mea-sures of climate risk exposure in different asset classes, and in particular for equityassets. Over the coming years, increased disclosure by firms — whether mandatedby regulators or demanded by large investors — will provide new opportunities tomeasure firms’ exposure to various types of climate risks. In the absence of newdata disclosed directly by firms, more creative use of already existing data — such assatellite imagery or text from 10-K statements or earnings calls — can be processedto improve climate risk exposure measures. Similarly, more sophisticated sentimentanalysis can improve our measures of negative climate news, as well as our abil-ity to separately identify news about physical and transition risk. Taken together,these improvements will improve our ability to construct increasingly more effectiveclimate hedge portfolios.

Another important question is to explore the extent to which climate risk, throughits effect on asset prices, may affect financial stability. The answer to this question de-pends, to a large extent, on the degree of concentration of these risks in the portfoliosof financial institutions and investors. Measuring this concentration is an importantand valuable research agenda. To do this, we require better measures of asset-levelrisk exposures, which could then be aggregated to the portfolio level. These numberswould allow financial institutions to better manage their climate risk exposures, andregulators to ensure that these risks do not pose a threat to financial stability.

ACKNOWLEDGMENTS

Stroebel acknowledges financial support from NBIM through their grant to theVolatility and Risk Institute at NYU Stern.

24

References

Alok S, Kumar N, Wermers R. 2020. Do fund managers misestimate climatic disasterrisk. The Review of Financial Studies 33:1146–1183

Arrow K, Cropper M, Gollier C, Groom B, Heal G, et al. 2013. Determining benefitsand costs for future generations. Science 341:349–350

Arrow KJ. 1995. Discounting climate change: Planning for an uncertain future.Lecture given at Institut d’Economie Industrielle, Universite des Sciences Sociales,Toulouse 24

Baker M, Bergstresser D, Serafeim G, Wurgler J. 2018. Financing the response toclimate change: The pricing and ownership of us green bonds. Tech. rep., NationalBureau of Economic Research

Bakkensen LA, Barrage L. 2017. Flood risk belief heterogeneity and coastal homeprice dynamics: Going under water? Tech. rep., National Bureau of EconomicResearch

Baldauf M, Garlappi L, Yannelis C. 2020. Does climate change affect real estateprices? only if you believe in it. The Review of Financial Studies 33:1256–1295

Bansal R, Kiku D, Ochoa M. 2019. Climate change risk

Bansal R, Ochoa M, Kiku D. 2016. Climate change and growth risks. Tech. rep.,National Bureau of Economic Research

Barnett M. 2017. Climate change and uncertainty: An asset pricing perspective.Tech. rep., Working Paper

Barnett M. 2020. Estimated impacts of climate policy risk. Tech. rep., Working Paper

Barnett M, Brock W, Hansen LP. 2020. Pricing uncertainty induced by climatechange. The Review of Financial Studies 33:1024–1066

Barreca A, Clay K, Deschenes O, Greenstone M, Shapiro JS. 2016. Adapting to cli-mate change: The remarkable decline in the us temperature-mortality relationshipover the twentieth century. Journal of Political Economy 124:105–159

Barro RJ. 2013. Environmental protection, rare disasters, and discount rates. Work-ing Paper 19258, National Bureau of Economic Research

25

Bernstein A, Gustafson MT, Lewis R. 2019. Disaster on the horizon: The price effectof sea level rise. Journal of financial economics 134:253–272

Binsbergen Jv, Brandt M, Koijen R. 2012. On the timing and pricing of dividends.American Economic Review 102:1596–1618

Black F, Scholes M. 1973. The pricing of options and corporate liabilities. Journalof political economy 81:637–654

Bolton P, Kacperczyk M. 2020. Do investors care about carbon risk? Tech. rep.,National Bureau of Economic Research

Brock WA, Hansen LP. 2018. Wrestling with uncertainty in climate economic models.University of Chicago, Becker Friedman Institute for Economics Working Paper

Brohe A, Greenstone M. 2007. The economic impacts of climate change: evidencefrom agricultural output and random fluctuations in weather. The American Eco-nomic Review 97:354–385

Burke M, Emerick K. 2016. Adaptation to climate change: Evidence from us agri-culture. American Economic Journal: Economic Policy 8:106–40

Choi D, Gao Z, Jiang W. 2020. Attention to global warming. The Review of FinancialStudies 33:1112–1145

Corelogic. 2019. The 2019 wildfire risk report

Crost B, Traeger CP. 2014. Optimal co2 mitigation under damage risk valuation.Nature Climate Change

Daniel KD, Litterman RB, Wagner G. 2019. Declining co2 price paths. Proceedingsof the National Academy of Sciences 116:20886–20891

Deschenes O, Greenstone M. 2011. Climate change, mortality, and adaptation: Evi-dence from annual fluctuations in weather in the us. American Economic Journal:Applied Economics 3:152–185

Desmet K, Rossi-Hansberg E. 2015. On the spatial economic impact of global warm-ing. Journal of Urban Economics 88:16 – 37

Dietz S, Gollier C, Kessler L. 2018. The climate beta. Journal of EnvironmentalEconomics and Management 87:258–274

26

Eichholtz P, Steiner E, Yonder E. 2019. Where, when, and how do sophisticatedinvestors respond to flood risk? Available at SSRN 3206257

Ellsberg D. 1961. Risk, ambiguity, and the savage axioms. The quarterly journal ofeconomics :643–669

Engle RF, Giglio S, Kelly B, Lee H, Stroebel J. 2020. Hedging climate change news.The Review of Financial Studies 33:1184–1216

Garnache C, Guilfoos T. 2019. A city on fire? effect of salience on risk perceptions.Tech. rep., working paper

Gibson M, Mullins JT, Hill A. 2017. Climate change, flood risk, and property values:Evidence from new york city. Tech. rep., Working Paper

Giglio S, Maggiori M, Rao K, Stroebel J, Weber A. 2020. Climate change and long-run discount rates: Evidence from real estate

Giglio S, Maggiori M, Stroebel J. 2015. Very long-run discount rates. The QuarterlyJournal of Economics 130:1–53

Goldsmith-Pinkham PS, Gustafson M, Lewis R, Schwert M. 2019. Sea level rise andmunicipal bond yields. Available at SSRN

Gollier C. 2002. Discounting an uncertain future. Journal of public economics 85:149–166

Gollier C, Weitzman ML. 2010. How should the distant future be discounted whendiscount rates are uncertain? Economics Letters 107:350–353

Golosov M, Hassler J, Krusell P, Tsyvinski A. 2014. Optimal taxes on fossil fuel ingeneral equilibrium. Econometrica 82:41–88

Hansen LP. 2014. Uncertainty outside and inside economic models. Journal of Po-litical Economy 122:945–87

Hansen LP, Miao J. 2018. Aversion to ambiguity and model misspecification in dy-namic stochastic environments. Proceedings of the National Academy of Sciences115:9163–9168

Hauer ME, Evans JM, Mishra DR. 2016. Millions projected to be at risk from sea-level rise in the continental united states. Nature Climate Change 6:691–695

27

Heal G, Millner A. 2013. Uncertainty and decision in climate change economics. Tech.rep., National Bureau of Economic Research

Hong H, Li FW, Xu J. 2019. Climate risks and market efficiency. Journal of Econo-metrics 208:265–281

Hsu PH, Li K, Tsou CY. 2020. The pollution premium. Available at SSRN 3578215

Huynh T, Xia Y. 2020. Climate change news risk and corporate bond returns. Journalof Financial and Quantitative Analysis, forthcoming

Ilhan E, Sautner Z, Vilkov G. 2019. Carbon tail risk. Available at SSRN 3204420

Issler P, Stanton R, Vergara-Alert C, Wallace N. 2019. Mortgage markets withclimate-change risk: Evidence from wildfires in california. Available at SSRN3511843

Johannes M, Lochstoer LA, Mou Y. 2016. Learning about consumption dynamics.The Journal of finance 71:551–600

Keenan JM, Bradt JT. 2020. Underwaterwriting: from theory to empiricism in re-gional mortgage markets in the us. Climatic Change :1–25

Kelly DL, Kolstad CD. 1999. Bayesian learning, growth, and pollution. Journal ofeconomic dynamics and control 23:491–518

Kolstad CD. 1992. Looking vs. leaping: the timing of co2 control in the face ofuncertainty and learning. Kaya et al 231

Krueger P, Sautner Z, Starks LT. 2020. The importance of climate risks for institu-tional investors. The Review of Financial Studies 33:1067–1111

Kurlat P, Stroebel J. 2015. Testing for information asymmetries in real estate mar-kets. The Review of Financial Studies 28:2429–2461

Lemoine D. 2015. The climate risk premium. Available at SSRN 2560031

Lemoine D. 2020. The climate risk premium: how uncertainty affects the social costof carbon. University of Arizona Department of Economics Working Paper :15–01

Lemoine DM, Traeger CP. 2012. Tipping points and ambiguity in the economics ofclimate change. Tech. rep., National Bureau of Economic Research

28

Li Q, Shan H, Tang Y, Yao V. 2020. Corporate climate risk: Measurements andresponses. Available at SSRN 3508497

Litterman R, Anderson CE, Bullard N, Caldecott B, Cheung ML, et al. 2020. Man-aging climate risk in the us financial system

Manne AS, Manne AS, Richels RG, Richels RG. 1992. Buying greenhouse insurance:the economic costs of carbon dioxide emission limits. MIT press

McCoy SJ, Walsh RP. 2018. Wildfire risk, salience & housing demand. Journal ofEnvironmental Economics and Management 91:203–228

Merton RC. 1973. Theory of rational option pricing. The Bell Journal of economicsand management science :141–183

Murfin J, Spiegel M. 2018. Is the risk of sea level rise capitalized in residential realestate. Working paper

Nordhaus WD. 1977. Economic growth and climate: the carbon dioxide problem.The American Economic Review 67:341–346

Nordhaus WD. 1991. To slow or not to slow: the economics of the greenhouse effect.The economic journal 101:920–937

Nordhaus WD. 1992. An optimal transition path for controlling greenhouse gases.Science 258:1315–1319

Nordhaus WD. 1994. Managing the global commons: the economics of climatechange, vol. 31. MIT press Cambridge, MA

Nordhaus WD. 2007. A review of the stern review on the economics of climate change.Journal of Economic Literature 45:686–702

Nordhaus WD. 2008. A question of balance: Weighing the options on global warmingpolicies. Yale University Press

Nordhaus WD, Boyer J. 2000. Warming the world: Economic models of global warm-ing. MIT press

Nordhaus WD, Popp D. 1997. What is the value of scientific knowledge? an appli-cation to global warming using the price model. The Energy Journal 18

29

Ortega F, Taspinar S. 2018. Rising sea levels and sinking property values: Hurricanesandy and new york’s housing market. Journal of Urban Economics 106:81–100

Ouazad A, Kahn ME. 2019. Mortgage finance in the face of rising climate risk. Tech.rep., National Bureau of Economic Research

Painter M. 2020. An inconvenient cost: The effects of climate change on municipalbonds. Journal of Financial Economics 135:468–482

Pedersen LH, Fitzgibbons S, Pomorski L. forthcoming. Responsible investing: Theesg-efficient frontier. Journal of Financial Economics

Perez-Gonzalez F, Yun H. 2013. Risk management and firm value: Evidence fromweather derivatives. The Journal of Finance 68:2143–2176

Piazzesi M, Schneider M, Stroebel J. 2020. Segmented housing search. AmericanEconomic Review 110:720–59

Rao K. 2017. Climate change and housing: Will a rising tide sink all homes? ZillowResearch

Sautner Z, van Lent L, Vilkov G, Zhang R. 2020. Firm-level climate change exposure.Available at SSRN 3642508

Stern N. 2007. The economics of climate change: the stern report. Cambridge, UK

Traeger CP. 2013. Discounting under uncertainty: Disentangling the weitzman andthe gollier effect. Journal of Environmental Economics and Management 66:573 –582

van Binsbergen J, Hueskes W, Koijen R, Vrugt E. 2013. Equity yields. Journal ofFinancial Economics 110:503 – 519

Van Binsbergen JH, Koijen RS. 2017. The term structure of returns: Facts andtheory. Journal of Financial Economics 124:1–21

Wagner G, Weitzman ML. 2015. Climate shock: The economic consequences of ahotter planet. Princeton, NJ: Princeton University Press

Weitzman ML. 2001. Gamma discounting. American Economic Review 91:260–271

Weitzman ML. 2007. A review of the stern review on the economics of climate change.Journal of Economic Literature 45:703–724

30

Weitzman ML. 2009. On modeling and interpreting the economics of catastrophicclimate change. The Review of Economics and Statistics 91:1–19

Weitzman ML. 2012. Rare disasters, tail-hedged investments, and risk-adjusted dis-count rates. Working Paper 18496, National Bureau of Economic Research

Weitzman ML. 2014. Fat tails and the social cost of carbon. The American EconomicReview 104:544–546

31